Does this move us closer to needing registries of authentic images? (White-listing? Only a partial solution.)
Google's response today, in addition to removing nano banana from Earth, was that people should just use Google's synthID(?) AI detection app, to determine if this is a blacklisted (effectively) image.
Of course that's impractical because a million people could quickly receive the image .. the image would make an impression on them before detection. And, this would also mean that Google's AI detection software needs to be ran by every human for every image that they view, in order to determine if it's valid. So, let's say that's 100 images, so that's a 100 to the millionth power? (Did I really get a CS degree?) I think it's just 100 million visits to Google synthID tool.. lol.
If we go with a white listing approach then we know that we should trust the image. If we go with a blacklisting approach then we only know that we should not trust the image, which we already knew before we ran the tool....
Watermarking is for fools. Those doing the generation and watermarking will control what is perceived as false or true. Watermarks have historically been about proof of ownership or origin. Watermarking fake creations works in reverse, all of a sudden what's void of watermarking is true, but those that control the watermarking can either add them to truths to turn them into fakes or skip watermarking for certain people or organizations, making their fake creations become truths.
There are a lot of things reasonable people would use AI for. This is not one of them. This is a crazy idea. Just plain bonkers. No sense to be found. "Oh, you want to see what's at that location? How about we make something up for you?"
What I find interesting - and seems absolutely logic for such a large company - that it's fine for a group of people to work on something tha is clearly to fail from Day 1, and it just seems to be ok.
I didn't use the feature, so I don't know if it was capable of this, but the thing you describe sounds potentially useful for land developers or folks trying to conceptualize buildings/city-development.
the people who make those stupid renders of yet-to-be-built strip malls filled with people and live bands that they use to sell development plans to city commerce groups.
When they talk about everybody losing their jobs, that doesn’t bother them one bit. In fact, they seem to be truly happy when they talk about that. But this, Oh No!
Is allowing a man to generate corn circles to send conspiracy theorists on a wild goose chase through the UK for the weekend really too much to ask for?
> Tonight I typed just one sentence into Google Earth and put refugees near the Mexican border. Then I planted a nuclear plant in Iran. What on earth is Google doing? Check my latest post here:
We're back to nerfing Google's products because "safety". Yes, you can obviously do this stuff, like you can obviously use Excel to open up a financial statement, change some numbers and tweet about it.
The obvious answer, and I believe it's already being utilized, is watermarks. There's been a lot of research in creating imperceptible, unremovable fingerprints from the generated image. Think very subtle statistical patterns in the generated pixels.
Instead we're going to get some long prompt about how not to do bad things that steps all over the user's intent.
Maybe LLMs will be the final nail in the coffin for putting every little ounce of every little article behind an annoying paywall.
A world model that ingests data in real-time from every flock camera, every iPhone, and every Rayban meta spy glasses won't have any need for journalists.
17 comments
[ 0.23 ms ] story [ 14.0 ms ] threadGoogle's response today, in addition to removing nano banana from Earth, was that people should just use Google's synthID(?) AI detection app, to determine if this is a blacklisted (effectively) image.
Of course that's impractical because a million people could quickly receive the image .. the image would make an impression on them before detection. And, this would also mean that Google's AI detection software needs to be ran by every human for every image that they view, in order to determine if it's valid. So, let's say that's 100 images, so that's a 100 to the millionth power? (Did I really get a CS degree?) I think it's just 100 million visits to Google synthID tool.. lol.
If we go with a white listing approach then we know that we should trust the image. If we go with a blacklisting approach then we only know that we should not trust the image, which we already knew before we ran the tool....
- https://www.digitaldigging.org/p/how-to-plant-a-nuclear-plan... original article is more interesting, also linked in above main article
the people who make those stupid renders of yet-to-be-built strip malls filled with people and live bands that they use to sell development plans to city commerce groups.
Google Earth's New AI Lets Anyone Fabricate Satellite Images
https://news.ycombinator.com/item?id=49124898
Is allowing a man to generate corn circles to send conspiracy theorists on a wild goose chase through the UK for the weekend really too much to ask for?
We're back to nerfing Google's products because "safety". Yes, you can obviously do this stuff, like you can obviously use Excel to open up a financial statement, change some numbers and tweet about it.
The obvious answer, and I believe it's already being utilized, is watermarks. There's been a lot of research in creating imperceptible, unremovable fingerprints from the generated image. Think very subtle statistical patterns in the generated pixels.
Instead we're going to get some long prompt about how not to do bad things that steps all over the user's intent.
https://x.com/henkvaness/status/2082912544394498228?s=20
A world model that ingests data in real-time from every flock camera, every iPhone, and every Rayban meta spy glasses won't have any need for journalists.